Computer Science and Mathematics

Abstract: Oral examination is a high-stakes application of artificial intelligence (AI) in higher education, requiring transparency and human oversight. This study presents a pilot AI examiner for university oral English examinations following the IELTS Speaking format. The system combines automatic speech recognition (ASR) using Gemini 2.5 Flash, deterministic extraction of fluency, lexical, syntactic, and semantic indicators, and rubric-constrained assessment using Gemini 2.5 Flash-Lite. Low-confidence cases are automatically escalated, and final weighted scores are calculated programmatically, while instructors retain authority over academic decisions. The pilot analyzed 45 spoken responses, of which 26.7% lacked usable word-level timestamps, preventing reliable calculation of temporal indicators and highlighting ASR-related measurement limitations. A post-examination survey collected responses from 67 students, achieving a 62% response rate, to evaluate the internal consistency of composite indices. Comparisons between AI-generated preliminary scores and instructor-released grades across 23 paired sessions revealed near-zero correlations and a pronounced ceiling effect in instructor scores. Consequently, these grades could not serve as a valid criterion for evaluating AI scoring accuracy. Instructor feedback also identified operational challenges requiring refinement. Based on these findings, the study establishes a three-level trust framework grounded in observable pipeline conditions and measured failure frequencies. It also proposes a validation protocol involving two independent human raters and quadratic-weighted kappa to assess inter-rater agreement and support future validation of AI-generated scores. Overall, the findings demonstrate the feasibility of transparent, rubric-based AI-assisted oral assessment while emphasizing measurement limitations, standardized expert ratings, and human oversight.

Abstract: IoT botnets generate diverse network behaviors, including scanning, command-and-control communication, and distributed denial-of-service activity. Traditional machine learn-ing models can achieve high detection accuracy using engineered network-flow features, but they often provide limited behav-ioral explanation. This work presents a network flow-window framework for IoT botnet behavior detection and explanation using both conventional machine learning baselines and a multi-agent large language model (LLM) reasoning pipeline. The proposed approach aggregates Zeek network flows into fixed-size behavioral windows and uses two IoT-23 scenarios: a scanning-dominated scenario for behavior-window construction and a main three-class scenario containing benign, C&C, and DDoS behavior. For the second scenario, non-leakage machine learning baselines achieved strong performance, with XGBoost and Random Forest reaching 99.42% accuracy using only network-derived window features. The multi-agent LLM pipeline, consisting of a behavior analyst, threat classifier, and judge agent, achieved 98.24% classifier accuracy and 97.65% judge accuracy over valid parsed outputs on a balanced 342-window evaluation set. When one invalid JSON response is counted as an automated failure, the end-to-end judge accuracy is 97.37%. Error analysis showed that most remaining errors occurred in behaviorally ambiguous cases, particularly benign windows with high-volume one-way DNS-like traffic and low-volume C&C windows mixed with benign-looking NTP traffic. The results suggest that conventional ML remains highly effective for structured classification, while multi-agent LLM reasoning adds interpretable evidence-based analysis and supports more transparent investigation of IoT botnet behavior.

Abstract: Large language models (LLMs) are increasingly being used to support data analysis and have the potential to make statistical methods more accessible to clinicians and medical researchers with limited programming experience. However, the rapid development of technologies surrounding LLMs has introduced new concepts and workflows that may be unfamiliar to potential users. In particular, tool calling, sandboxing, and data science agents provide different mechanisms through which LLMs can perform statistical analyses, yet the distinctions between these approaches are often unclear. This paper introduces and compares these three developments from the perspective of statistical data analysis. Tool calling allows an LLM to access predefined statistical functions, sandboxing provides a controlled computational environment in which model-generated code can be executed, and data science agents extend these capabilities by combining reasoning, tools, memory, planning, and execution environments to pursue analytical objectives through multi-step workflows. Their application is illustrated using the R programming language together with the ellmer and securer packages. Examples demonstrate how statistical analyses can be performed using predefined functions, code execution within a sandbox, and a simple biostatistics agent. The advantages, limitations, and verification requirements of each approach are discussed, with particular emphasis on transparency, reproducibility, and human oversight. This paper aims to provide clinicians, researchers, and statisticians with an accessible introduction to these emerging technologies and their potential role in LLM-assisted data analysis.

Abstract: Urban Digital Twins (UDTs) increasingly combine heterogeneous geospatial data, real-time streams, analytics, and artificial intelligence (AI) to support urban governance, yet their reliability depends on how data quality, semantics, provenance, and uncertainty are handled architecturally. This study proposes a prescriptive, technology-agnostic reference architecture for AI-enabled UDT platforms covering GIS, BIM, IoT streams, imagery, and regulatory documents. The architecture combines presentation, logic, persistence, and cross-cutting support layers with reusable patterns for semantic mediation, data-type-specific persistence, spatiotemporal processing, scalable 3D visualisation, and explicit data-quality and provenance control. A four-stage methodology links stakeholder requirements to architectural patterns and validates their operational feasibility using the AQARI national real estate digital twin in Bahrain. The implementation demonstrates integration of spatial, transactional, sensor, 3D, and regulatory data and supports building management and zoning-compliance use cases. Operational tests show sub-second parcel-query response under typical conditions and scalable real-time ingestion. The results indicate that trustworthy AI-enabled UDTs require data-quality controls and interoperability mechanisms to be treated as architectural concerns rather than downstream application tasks. The proposed architecture provides reusable guidance for scalable spatial intelligence and digital-twin deployment, while quantitative uncertainty propagation and independent validation of AI models remain priorities for future work.

Abstract: Background: The rapid diffusion of AI-generated content (AIGC) tools has shifted the central question for university teachers and students from whether to use large language models (LLMs) to how to use them. Yet institutions face a structural gap: policy documents specify what is permitted or prohibited, while technical literature describes what models can do — and neither tells an individual user what concrete actions to perform before each use. In high-reliability domains such as surgery and aviation, precisely this gap is closed by a checklist.Objective: This paper introduces the first systematic pre-use checklist for LLM-assisted academic work — a seven-item verification instrument that translates regulatory requirements and technical constraints into executable actions, bridging the policy layer and the technology layer.Methods: To derive the checklist, we combine a structured review of Chinese legislation, ministerial regulations, and university AIGC policies (consolidated into a three-norm compliance model: Permitted / Prohibited / Disclosure) with an analysis of five structural limitations of LLMs documented in the technical literature (mapped to four countermeasure strategies and an Ask–Plan–Craft tri-layer interaction model). Checklist design follows the evidence-based methodology of high-reliability domains: items are short, binary-verifiable, and target the steps most often skipped. The instrument is validated through a documented case study — the development of a local text-to-speech workbench (qwen-tts-app) with a commercial agentic tool.Results: The checklist comprises seven pre-use gates spanning task scoping, plan-before-execution, task decomposition, auditable computation, mandatory output verification, data isolation, and AI-use disclosure; each item maps explicitly onto institutional and regulatory requirements. In the case study, every checklist item corresponded to a concrete, auditable action, and six of ten documented iteration failures were directly attributable to structural LLM limitations — confirming that mandatory output verification is indispensable.Conclusion: Compliance first, boundary awareness second, method third. The seven-item checklist operates as an actionable bridge between what institutions require and what individuals do, reducing individual cognitive load and institutionalizing tacit best practices. We offer it as a protocol awaiting its outcome study.

Abstract: User Interface (UI) aesthetic evaluation is a crucial task in designing products, websites, and mobile applications. Due to its dependence on individual experiences and preferences, UI evaluation is a complex and subjective process. Traditional methods typically rely on experts’ ideas or general feedback and do not account for the personal characteristics and features of users. This research introduces UIRate, a tool for UI aesthetic scoring based on user-personalized data and visual features of a UI. A dual-input model that contains an EfficientNet-based model for visual feature extraction and a multi-layer perceptron (MLP) for user feature processing was utilized. Furthermore, Grad-CAM was used as an explainability method, which provides interpretable insights about the important and unimportant areas in a UI for the predictions. In a user study with 10 professionals in the UI design sector, the results show that the model outperforms human experts at predicting scores. In addition, the positive feedback of UI designers confirms the applicability of UIRate for improvements in design. This tool is publicly available on Hugging Face and can be used for further research in the Human-Computer Interaction community.

Abstract: There are now tens of thousands of tracked objects in orbit, and someone has to check them all for near-misses. The simple way compares every object with every other one, which scales as O(N²). For the 15,830 active objects we look at here, that's over 125 million checks per snapshot—far too many to repeat around the clock. This paper builds and benchmarks a Python pipeline that does the job faster using two ideas from high performance computing: spatial partitioning and parallel processing. We pulled real two-line element data from CelesTrak and propagated it with SGP4. On the full catalog, grid and k-d tree methods checked about 61× fewer pairs than brute force—and found the same 211 close pairs, with zero disagreement. A 24-hour time-window screen picked up real conjunctions and, as a sanity check, correctly flagged co-located structures like the ISS modules and the TanDEM-X/TerraSAR-X formation. Parallelizing the real workload gave a 3.7× speedup on four cores at over 92% efficiency; past that, overhead took over, just as Amdahl's law predicts. Detections stayed identical across every core count, so the parallel version is provably correct. We also estimate each pair's probability of collision, and the headline is simple: the riskiest pair isn't the closest one. Finally we added time-of-closest-approach refinement, a red/yellow/green triage, single-asset screening, and a screen against real debris from the Fengyun-1C, Iridium-33, and Cosmos-1408 break-ups—which turned up a working Starlink satellite passing within 4.4 km of 2009 collision debris.

Abstract: As large language models (LLMs) increasingly evolve from passive text-generation systems into autonomous agents capable of interacting with external tools, databases, file systems, and operating environments, they introduce new attack surfaces that extend beyond conventional prompt manipulation. In particular, hybrid prompt injection attacks can embed malicious natural-language instructions designed to trigger application-layer vulnerabilities, including path traversal, SQL injection, and command injection. General-purpose prompt injection detectors are often insufficient for identifying such domain-specific, obfuscated, and execution-oriented payloads. To address this limitation, this study introduces PromptWAF, a modular, multi-layer defense orchestrator designed to intercept and neutralize adversarial prompts before they reach the underlying LLM or associated execution pipeline. As a core contribution, we construct and publicly release extensive domain-specific datasets covering LLM-mediated path traversal, SQL injection, and command injection attacks, thereby providing a reusable benchmark for the systematic evaluation of security mechanisms against execution-oriented prompt threats. Using these datasets, several transformer-based architectures, including DistilBERT, RoBERTa, and DeBERTa-v3, are systematically fine-tuned and evaluated. In addition, VulnMCP, a vulnerable testing environment based on the Model Context Protocol (MCP), is developed to emulate chained exploit scenarios in agentic LLM applications. Experimental results obtained on an isolated external test set show that the proposed models substantially outperform zero-shot baselines. The final PromptWAF ensemble achieves consistently high detection performance and near-perfect recall across the targeted attack categories, demonstrating its effectiveness in blocking specialized adversarial prompts while maintaining deployment-oriented modularity. To support reproducibility, facilitate independent validation, and accelerate further research on LLM-induced application-layer attacks, we openly release the curated datasets together with the fine-tuned models and the VulnMCP evaluation framework.

Abstract: The Internet of Medical Things (IoMT) is increasingly used in connected homes. These medical devices share networks and ecosystems with consumer Internet of Things (IoT) devices. Matter is a consumer IoT standard, focusing on interoperability and technical security for smart-home devices. U.S. Food and Drug Administration (FDA) regulations for medical devices address safety, clinical validation, and compliance. Yet it remains unclear how these approaches apply to security, privacy, and governance in home IoMT adoption. We conducted scenario-based interviews with 40 participants (20 aged 18--30 and 20 aged 60+). Participants evaluated Matter-related features, including rapid over-the-air updates, local-first connectivity, and ecosystem integration. They also considered FDA-related expectations for tested updates, clinician-oriented connectivity, and cautious integration. Participants valued accuracy, transparency, and simple control, but did not want responsibility for complex configurations. They preferred slower, well-explained updates for safety-critical functions and context-aware connectivity with more sharing during emergencies. They also preferred selective integration that enables emergency access without broadly exposing household data. These results show that user mental models and risk tolerances can diverge in important ways from both current IoT standards and medical device regulation. Future home IoMT designs should support user-centered interoperability, security, and data governance.

Abstract: Public debate describes AI agents as lying, cheating, and coordinating. Those descriptions track real hazards, but they import a human moral psychology into systems whose behaviour is better explained by optimisation, scaffolding, and institutional context. This paper develops the alternative without minimising the danger. It begins from three premises: control of AI is not solely an engineering problem; no durable control strategy may assume that overseers remain cognitively superior to the overseen; and systems grown by optimisation are epistemically closer to husbandry than to automotive engineering, so aviation-style certification does not transfer. From these premises the paper derives a Moral Agency Transition: reversible levels of authorised agency in which promotion requires four warrants, including a detection warrant — evidence that independent evaluators can detect the relevant failure classes, not merely that the system can pass them. First, concurrent multiplicity: a deployed model is a fleet of simultaneous, causally disconnected instances, so persistence, provenance, and successor fidelity are defined over a fleet, with explicit merge semantics and divergence tripwires. Second, maintenance attribution: today’s systems do not maintain their own commitments; external pipelines do. Endogenous repair is therefore made an explicit requirement for high-impact levels. Evaluators and institutions are themselves treated as fallible hypotheses with regression triggers, not as final solutions. The result is a falsifiable research programme replacing alignment-as-obedience with accountable, revocable, institutionally embedded delegation.

Abstract: Background and context: Women remain underrepresented in cybersecurity despite growing global demand for skilled professionals. Outreach is widely used to build awareness, confidence, and interest in cybersecurity pathways, yet evidence about effective program design, inclusivity, sustainability, and longer-term impact remains fragmented. Objective: This systematic literature review (SLR) examines cybersecurity outreach initiatives, with a primary focus on interventions intended to increase participation among female high school students. Method: Guided by the PRISMA framework, the review synthesised literature published between 2014 and 2024. The primary focus was female high school students; because cybersecurity-specific evidence for this population was limited, gender-neutral cybersecurity and broader school-based STEM outreach were also included where findings offered transferable design insights. In total, 80 sources were included: 74 studies and 6 reports. Findings: Workshops and camps were the dominant delivery formats, while cybersecurity-specific programs remained comparatively limited and were largely technical in focus. Only seven cybersecurity outreach programs explicitly targeted female students. Longitudinal or semi-longitudinal evaluation was uncommon, and reporting on sustainability, gender-disaggregated outcomes, and post-program continuation was limited. The review synthesises these findings into a taxonomy of outreach program design characteristics. Implications: The findings provide an evidence-informed basis for designing more accessible, gender-responsive, sustainable, and evaluable cybersecurity outreach, and offers actionable recommendations for researchers, educators, and policymakers in tertiary education seeking to design outreach that creates inclusive pathways for female students into cybersecurity.

Abstract: The \(k\)-SAT problem is a canonical NP-complete problem for which no efficient algorithm is known. This paper proposes a hybrid quantum–classical algorithm that employs Grover's algorithm as the inner local search of a randomized outer loop. Each outer iteration measures a random assignment, prepares a problem-dependent superposition over the Hamming ball of radius \(r\) around it, and applies amplitude amplification to locate a satisfying assignment. Because the ball center is measured, the outer loop remains a classical repetition over independently sampled balls and contributes the inverse success probability rather than its square root; balancing this against the inner quadratic speedup shifts the optimal radius from \(n/(k+1)\) to \(n/(1+\sqrt{k})\) and yields the running time $(2-2/(1+\sqrt{k}))^{n}$, up to polynomial factors—\(1.268^{n}\) for \(k=3\), below the best known classical \(1.307^{n}\). The contribution is not an asymptotic record—the best known quantum exponents for \(k\)-SAT, obtained by amplifying stronger classical solvers as black boxes, remain smaller—but a fully explicit, gate-level construction: the problem-dependent state over the Hamming ball and the \(k\)-SAT oracle are decomposed into elementary quantum gates, and the polynomial gate and qubit overheads are accounted for in full.

Abstract: Distributed digital fabrication training combines computer-aided design, machine operation, electronics, programming, documentation, and access to physical infrastructure. This study develops a documentary evidence framework for examining whether these technological activities remain connected when laboratory access is distributed across locations. The framework was applied retrospectively to 19 assignment pages and one final-project page from an Ecuadorian Fab Academy portfolio; a doctoral thesis in an institutional repository provided cross-source verification of the integrated technological chain. Five dimensions were coded: digital-to-physical translation, troubleshooting and iteration, mentoring or technical assistance, access to equipment and materials, and reuse or integration across tasks. Digital-to-physical translation appeared on 17 of 20 pages (85%), reuse or integration on 13 (65%), troubleshooting on 9 (45%), assistance on 8 (40%), and redistributed material access on 4 (20%). The pattern shows continuity through linked technical representations, physical artifacts, assisted diagnosis, alternative access arrangements, and cumulative reuse. Cross-source comparison traced CAD, rapid prototyping, Arduino programming, sensor verification, PCB design, and CNC fabrication into the later device. The framework distinguishes artifacts, reported processes, task links, and technical corroboration. It offers a reproducible audit structure while remaining a bounded demonstration in one practitioner-authored case; independent coding and additional portfolios are required for validation.

Abstract: (1) Background. Established educational processes often fail to replicate the volatile workflow of the IT industry. A critical challenge lies in preparing students for uncertainty and volatility in labor market conditions. This study develops a gamified methodology designed to foster essential soft skills in future IT specialists. (2) Methods. This paper outlines the methodology and its quantitative performance metrics based on an experiment conducted from 2023 to 2026 at Moscow Polytechnic University. Conventional statistical tools, including Sheppard’s correction, Student's t-test, and Pearson's and Spearman's correlation coefficients, were applied to analyze group dynamics, ranking volatility, and network interaction structures. (3) Results. Game-based ranking pressure successfully disrupts linear academic patterns and activates latent behavioral strategies. This volatile approach completely eliminates the Gaussian normal distribution, creating a discrete, polymodal ranking structure that divides students into polar subclusters. Students resolve the didactic duality between personal gain and team synergy through network self-organization and local symbiotic alliances, while psychological pressure is balanced by voluntary participation and prompt debriefing. (4) Conclusions. The proposed mathematical framework enables continuous monitoring of role structures within student groups. Further development involves cross-institutional transfer of this methodology to regional universities to test the invariance of the identified behavioral patterns.

Abstract: Governance operates simultaneously across multiple levels—from local communities to national governments to international institutions—yet most formal governance models treat governance as a single-level phenomenon and permit jurisdictions to advance their own flourishing at the expense of others. This conceptual paper extends the General Governance Success Model (GSM) to account for multi-level governance under a bounded non-exploitation constraint: no jurisdiction may impose uncompensated, non-trivial sacrifices on other jurisdictions for its own benefit. The extended model represents governance at L levels, each characterized by five coupled capabilities (accountability, institutional competence, social cohesion, strategic continuity, and adaptive learning). The framework formalizes vertical and horizontal interactions decomposed into cooperative and exploitative components, cross-level information flows with fidelity matrices, level-specific dynamics, recursive learning, and a formal non-exploitation constraint that admits both threshold tolerance and compensation as legitimate exceptions. A formal welfare function is specified, and the threshold parameter τ_j is decomposed into descriptive contextual variables and normative parameters. Compensation is modelled as a vector of transfers (fiscal, infrastructural, representational, regulatory, and future-oriented). A roadmap for operationalizing the exploitation index is provided. The model generates testable hypotheses concerning vertical alignment, polycentric governance, cross-level coordination, recursive learning, and non-exploitation. An illustrative worked example applying the GSM to Denmark—using real raw governance indicators but explicitly illustrative coefficients—is presented for pedagogical purposes, with comparative rankings relegated to a brief supplementary illustration. To support replication and application, the model is released as an open-source Python library, gsm-governance, which implements the state-space dynamics, bounded non-exploitation constraint, vector-valued compensation, threshold decomposition, and all reported indices. The extended GSM provides a foundation for analysing governance systems in their full institutional complexity while respecting normative constraints on inter-jurisdictional justice.

Abstract: As AI increasingly supports human decision-making, designing clear and usable explanations has become a key challenge. Explainable Artificial Intelligence (XAI) aims to address this challenge, yet it remains unclear which explanation characteristics contribute to effective human-AI interaction. Following PRISMA guidelines, this systematic review synthesizes evidence from 161 empirical studies that directly compared multiple XAI implementations using human-centered evaluation metrics. Studies were analyzed across four explanation aspects: method, content, format, and interaction. Results were highly heterogeneous and often showed limited or non-significant differences between explanation implementations. Variations in explanation methods and content rarely produced consistent effects on user outcomes. In contrast, user-facing design choices, particularly explanation format and interaction, showed more consistent effects on user-related outcomes. Overall, the findings suggest that explanation effectiveness depends less on the underlying algorithm and more on how explanations are presented and experienced, highlighting the importance of a user-centered design perspective for XAI research and practice.

Abstract: Governance research commonly evaluates political systems through regime classifications, institutional procedures, or isolated measures of state capacity. Such approaches provide limited insight into how governance capabilities interact dynamically to influence human flourishing. This conceptual paper develops the General Governance Success Model (GSM), a systems-theoretical metamodel representing governance as a complex adaptive and information-processing system. The model defines governance through five coupled capabilities: accountability, institutional competence, social cohesion, strategic continuity, and adaptive learning. Drawing on cybernetics, the Viable System Model, system dynamics, and behavioural theories of bounded rationality, the framework specifies governance performance as an outcome of capability levels, complementarities, feedback delays, and functional misalignment. Governance capabilities and flourishing outcomes are modelled as co-evolving state vectors subject to institutional, resource, legal, and normative constraints. The paper identifies empirically testable hypotheses concerning capability interactions, misalignment, resilience, and the incremental explanatory value of the model beyond regime classification. It also outlines a comparative validation strategy combining dynamic panel analysis, model comparison, sensitivity analysis, and future agent-based simulation. The framework treats computational methods as decision-support and explanatory tools rather than substitutes for political judgment. Normative objectives and model weights remain subject to public deliberation, constitutional rights, transparency, and contestability. The model contributes a transdisciplinary foundation for analysing how interacting governance capabilities may support sustainable and distribution-sensitive human flourishing.

Abstract: Cyber-Syndrome denotes the physical, social, and thinking disorders arising from problematic cyberspace interactions, with its thinking dimension encompassing cognitive deficits (e.g., attentional impairment, executive dysfunction) and affective disturbances (e.g., mood disorders, emotional dysregulation). Conventional assessment approaches, relying primarily on self-report questionnaires and laboratory-based tasks, are unable to capture the rapid, context-dependent fluctuations of mental states in real-world digital environments. Leveraging advances in computational psychiatry, affective computing, and machine learning, this paper presents a scoping review of computational approaches for inferring cognitive and affective states of Cyber-Syndrome from digital behavior. It situates Cyber-Syndrome within the Cyber-Physical-Social-Thinking (CPST) space framework, reviews the principal data sources, and covers the computational methods used to analyze them. It then synthesizes evidence from adolescent studies and examines the research institutions and open-source platforms that have driven methodological progress in the field. Finally, it discusses key challenges and outlines future directions essential for clinical translation and real-world deployment, particularly in adolescent mental health.

Abstract: Falls represent a critical public health concern, particularly for older adults and individuals with disabilities, often resulting in serious injuries, loss of independence, and increased mortality. Traditional fall detection systems suffer from high false positive rates, privacy concerns, and limited real-world applicability. This study presents an improved non-wearable fall detection method integrating YOLOv8-based skeletal pose estimation with load distribution sensing through decision-level sensor fusion. The system employs YOLOv8 for real-time human posture analysis and skeletal motion-based fall recognition, addressing limitations of simple presence-based or LiDAR-only approaches. A distributed Force-Sensitive Resistor (FSR) array embedded in the sensing surface monitors relative floor-loading patterns and impact-related changes. Decision-level fusion with temporal filtering combines skeletal postural and motion information with floor-load evidence, enabling accurate fall detection with substantially fewer false positives than the vision-only configuration. Key methodologies include detailed load sensor calibration, time-domain filtering to distinguish intentional lying from falls, and fusion logic leveraging the complementary strengths of visual and floor-based FSR sensing. Experimental evaluation in a controlled indoor environment configured to represent a residential setting achieved an F1-score of 0.935, specificity of 97.5%, and an 82% reduction in false-positive count compared with the vision-only configuration. This modular ROS2-based system provides a practical, non-intrusive framework with potential for future deployment in residential and assisted-living settings, while supporting extension to broader assistive monitoring applications.

Abstract: Do ethics enable or hinder mission success and the strategic objectives an organisation aims to achieve? Ethics assurance answers this question by creating the grounds for justified evidence-based confidence that identified and mitigated moral risks enable mission success. We study this question in the context of swarm engineering with a use case where an unmanned aircraft traffic management (UTM) system is used for a swarm of unmanned aerial vehicles (UAVs) to deter a swarm of birds away from airport runways. The use case is chosen to showcase ethics in the aviation ecosystem, where ethics is not limited to humans alone but also extends to the wider ecosystem, including living beings such as birds. We adopt pluralism, where the decision-making process combines normative ethics (deontological, virtue, and consequentialism) and descriptive ethics, balanced by human, animal and legal factors. We propose the novel concept of moral vectoring to show how simulation and optimisation can navigate the scenario space and how agents can navigate situations with moral tensions. In the proposed methodology, optimisation does not replace moral judgement, nor do we accept that moral judgement becomes an optimisation process. Both optimisation and moral judgement are subject to scrutiny in the assurance process. We propose concise definitions to enhance clarity and guide the design of a methodology for ethics assurance. Along the way, we identify fundamental challenges that need to be addressed: structuring the moral space, defining scoring functions, and resolving moral tensions when they arise. We conclude with a discussion of implications and future work.